CHANGES
=======

* inelegant solution by adjusting the list comprehension in line#252
* Typos and discord community url fix

v0.2.1
------

* Added discord and documentation urls
* tutorial notebook updated
* added tests for monitoring
* test for cropping fixed + samna requirement bump
* DVSLayer.from\_layers take an input of len 3 Added checks for input\_shape
* ci updated to not wait for confirmation
* replaced swapaxes with transpose for backward compatibility with pytorch 1.7
* gitlab ci updated to install latest version of samna
* added doc strings
* Added instructions for how to add support for a new chip
* api docs nested
* wip
* deleted mnist\_speck2b example script as dynapcnn\_devkit works by just replacing the device name
* update API doc
* Default monitor enabled for last layer is nothing is specified
* merged changes
* Removed redundant/legacy code
* rename API doc headings
* Update unit tests according to recent commits
* clean up API documentation by not displaying module names
* Minor fixes and adaptations. More specific exception type. Can pass network with dvs layer to dynapcnn compatible network
* Smaller fixes in config dicts
* Refactored dvs unit tests
* fixed typos in documentation
* Bug fix in crop2d layer handling
* Added Crop2d layer
* installation instructions and minor documentation changes
* added some folders to gitignore
* moved event generation methods to ChipFactory
* depricated methods deleted from source
* supported\_devices in ChipFactory and get\_output\_buffer in ChipBuilder
* added support for time-stamp management
* enable pixel array on dvs\_input true
* adding speck2b device names + mnist example script
* speck2b bug fix in builder
* removed factory setting line
* added speck2b to the condition
* added speck2b to the condition
* added parameter file for example
* Added config builders for speck and speck2b
* Refractored to add ConfigBuilder
* Support for InputLayer. Still does not pass \`test\_dvs\_input\` tests
* added index for samna
* Cut dimensions update in the configuration dict
* Minor api corrections
* dynapcnn layers populaiton works. Bug in dvs layer still to be sorted out
* wip: build full network
* construct dvs layer construction works
* Added tests for DVSLayer
* Added custom exceptions and tests
* method to build network fo dynapcnn layers added
* added start layer index to construction methods
* Added tests for layer builders
* Added function to create dynapcnn layers
* DVSLayer, FlipDims functional code added
* Suggestion: DVSLayer. Still to be completed
* WIP
* Added handling of sumpool layers at the start of the model
* Updated MNIST example notebook in the documentaion
* added speck2\_constraints
* make\_config default for chip\_layers\_ordering changed to "auto"
* unhide chip\_layers\_ordering
* Breaking change: monitor\_layers now takes model index instead of chip layer index
* wip
* Added API docs for new files
* Added the basic documentation
* doc skeleton added for the fundamentals
* mapping logic updated to edmond algorithm
* bug fix in make\_config effecting auto-mapping
* Layer dimensions infereed from dimensions dict
* updated memory summary to take chip constraints
* samna warning message raised
* open device checks if the device is already open
* moved monitor to make\_config
* added xytp conversion methods
* added warning for discretization
* added test for auto in make\_config
* Added timestamping and memory\_summary methods
* Bug fix: Padding and stride x, y swapped
* Events to raster marked as NotImplemented
* Time stamped events generated
* Forward method defined on events
* Bug fix: config invalid when network uninitialized (no data passed)
* added bug fix for str 'speck2devkit'
* Added option to specify which layers to monitor in to method
* to device method implemented
* samna device discovery memory errors fixed
* get\_opened\_devices also returns device\_info object
* added get\_device\_map
* Added device\_list
* Added method to discover connected devices
* wip: find/move model to device when to() is called
* Config object conditionally created based on device type
* added further test
* Correct error now raised if spiking layer missing at end of network
* Added io file
* Raise warning when discretize is True and there is an avgpooling layer
* Revert "need to test if samna is there"
* membrane reset now implemented properly
* pypi deploy new line added
* sphinx requirements added
* typo in conf.py fixed
* docs folder relocated
* setuptools based setup file
* pbr based project versioning and gitlab ci added
* Samna requirement updated
* Method parameter in test corrected
* changed speck to dynapcnn
* fixed mapping problem in auto layer order
* Replace all references to speck as DYNAPCNN, including internal variables
* Type annotation fixed
* Refractored code to dynapcnn from speck
* fixed bug in discretization of membrane\_subtract (double multiplication)
* updated tests to reflect changes in sinabs network

v0.2.0
------

* fixed docs, removed commented-out areas
* removed dependency on samna for validation, and on SpikingLayerBPTT
* Changed 'input\_layer' management for sinabs changes'
* fixed tests, one not passing
* started changing dvs\_input default
* added dropout
* Roll back changes from last commit and only make sure that meaningful error is produced when last layer is not spiking. Handling of last layer done in sinabs from\_model
* wip: handle networks that end with linear or conv layer
* fixed true\_divide torch fussiness
* removed print statement
* merged commit with sumpool support
* implemented support for sumpool in input network
* Disable default monitor and support one dvs input channel
* In-code docs for test\_discretized
* Smaller fixes in discretize
* Tests for discretization module
* Added leak management, and test
* individual\_tests made deterministic
* fixed input tests
* valid\_mapping complies with variable naming convention. Extended in-code documentation
* Minor fix in test\_dvs\_input
* Ignore jupyter checkpoints
* Placeholder in tutorial for validation and upload to Speck
* Fixes in test\_dvs\_input
* Rename test\_dvs to test\_dvs\_input
* test\_dvs: Tests with input\_layers
* Warn if both input\_shape and input layer are provided and shapes don't match
* test\_dvs: make sure that missing input specifications are detected
* test made deterministic
* Removed requirement of samna, particularly for tests
* added skip tests with no samna
* doorbell test fixed
* updated large net test to an actual test
* added tests; added support for 3d states
* fixed bug DVS input size
* extended tests to config
* deal with missing neuron states
* automatic choice of layer ordering
* add handling swapping layers while searching for a solution
* removed prints, fixed test
* Many fixes needed for the configuration to be valid. Now works
* Documentation for discretize
* Cannot change conv and spk layers, but access them through property. Pool can be changed
* Cannot change conv and spk layers, but access them through property. Pool can be changed
* getting closer
* improvements
* working check on real samna
* validation thing to be compared across machines
* Specklayer correctly handles changing layers. Todo: Update unit tests
* wip: specklayer: make sure that when changing layers, config dict gets updated. TODO: unit test fails
* Property-like behavior for conv/pool/spk layers
* Comparison with original snn only when not discretizing
* Ensure no overwrite of the conv layer during batchnorm merging
* Making sure discretization happens after scaling
* Tutorial for converting from torch model to speck config
* Update documentation
* WIP: Documentation for specklayer. Numpy style docstrings
* WIP: Sphinx documentation
* Minor fixes. Still to do: Discretization of snn (discretize\_sl) does not work)
* Minor fixes in tests
* fixed bug in sumpool config
* Fixed SumPool
* Completed name change and move of files
* Fix module naming
* deleted references to sumpool2dlayer, loaded sinabs sumpool
* removed unused imports
* uses SumPool from sinabs
* moved test
* updated tests to new locations; new constructor in SpeckNetwork
* moved tests to folder
* deleted scratch folder
* Tests related to dvs
* Fixes wrt to handling dvs and pooling, completed type hints
* wrote docstrings
* should now be safe to commit init
* some minor changes
* added test, changed var names
* small correction to previous commit
* added support for a specific case of batchnorm
* Use deepcopy for copying layers
* merge bc of black
* Avg pooling now turned to sum pooling and weights rescaling (1 failing test)
* Test to verify that all layers are copy and not references
* Make sure all layers in SpeckCompatibleNetwork are copies of the original
* (WIP) started implementing transfer to sumpool
* Workaround for copying spiking layers in discretize\_conv\_spike
* updated and added tests
* fixed several issues that arose with testing
* correct way of ignoring neurons states
* discretization now optional (for testing)
* input shape removed where not needed; more cleanup
* Minor
* separated make\_config from the rest
* a little cleanup and commenting
* seemingly working class-based version
* somewhat working version of class-based
* Handle Linear layers and Flatten, ignore Dropout2d
* started transformation into class
* added gitignore
* updated new api of samna
* added smartdoor test
* Doorbell test
* Un-comment speck related lines
* minor
* samna independent test-mode for fixing some bugs
* Fixing bugs
* Wip: update for sinabs 0.2 - discretization
* Wip: update tospeck for compatibility for sinabs 0.2
* Wip: update tospeck for compatibility for sinabs 0.2
* removed content from init file, since it breaks for people who do not have dependencies
* introduced test of discretization in simulation
* Made necessary changes to actually simulate the discretized network
* added bias check in descretize sc2d
* merged changes from features/separate\_discretization
* bugfixes
* fix import
* misc
* Fix bias shapes and handling of 'None'-states
* merge updates from feature/spiking\_model\_to\_speck
* Fix biases shape
* wip
* Fixes in neuron states and weight shapes. Updated test
* Undo reverting commit 5af49846 and fix dimensions for neuron states
* Fix weight dimensions
* Use speckdemo module and handle models without biases
* Small fix in plotting in test
* Improved in-code documentation of tospeck.py
* Test script for porting a simple spiking model to a speck config
* Quantization of weights, biases and thresholds
* Bugfixes in tospeck.py
* can handle sequential models of SpikingConv2dLayers and SumPooling2dLayers
* Remove tests that should be handled by ctxctl
* wip: handling of pooling layers
* For compatibility issues that result in not matching dimensions raise exceptions instead of warnings
* WIP: Method for converting Spiking Model to speck configurations
* WIP: Method for converting SpikingConv2DLayer to speck configurations

0.1.dev7
--------

* initial mock code
* Initial commit
